기계공학부 구성원들의 많은 관심과 참여 부탁드립니다.

▣ 주   제: Advanced System Integration Reliability in HPC and AI Network Systems: A Landscape

▣ 연   사: 이태규 박사

▣ 소   속: Cisco Systems

▣ 일   시: 2025. 12. 12.(금) 16:00

▣ 장   소: 제4공학관 D403호

▣ 초   록

Advances in high-performance computing (HPC), and AI network systems are redefining requirements for electronic interconnects. Modern architectures demand concurrent scale-down, scale-up, and scale-out strategies, each imposing distinct challenges in materials, manufacturing, and reliability. At the device level, heterogeneous integration and chiplet-based partitioning are driving finer interconnect pitches and larger die and larger body size footprints. The continued adoption of hybrid Cu-to-Cu bonding underscores copper’s central role, as its electrical performance and process compatibility remain essential for dense on-chip and die-to-die communication. At the system level, scale-out architectures linking ASICs, accelerators, and switches require significantly increased power delivery. While Joule-heat-induced temperature rise at individual interconnects may appear limited, cumulative thermal effects influence resistance drift, aging behavior, and signal integrity. These trends have elevated interest in Co-Packaged Optics (CPO) to mitigate bandwidth and energy constraints, and in large organic or glass substrates to accommodate expanding routing demands. Reliability under these conditions is strongly governed by phase-transformation phenomena, including melting, solidification, and creep. Simultaneously, extreme mechanical and thermal environments, ranging from automotive vibration to cryogenic quantum systems and high-temperature edge robotics, reaffirm the importance of interconnect integrity and system integration. This paper presents a unified assessment showing that fundamental metallurgy and classical mechanical engineering including copper and solder joint interconnects, phase-transformation, and multi-physics, remain foundational to next-generation system design in high-performance computing (HPC), and AI network systems.